MiMo V2 TTS
MiMo V2 TTS was Xiaomi's speech-generation model before the V2.5 TTS series. It combined a proprietary audio tokenizer with joint speech-text modeling and reinforcement learning.
Retired
Written cues changed the spoken performance
The release described global style directions and local controls for laughter, pauses and other vocal events. Capital letters and repeated words could affect emphasis and rhythm. Singing was part of this speech model, rather than evidence of a separate instrumental-music model.Original MiMo V2 TTS announcement
The original API name has retired
The original API name expired on 30 June 2026 at 00:00 UTC+8. Xiaomi listed V2.5 TTS as its replacement.MiMo model retirement and replacement schedule
Practical guidance. This page records the original release. A request routed to a successor does not establish that the original model is still being served. Use the current model list when choosing an API.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Retired |
| Launch date | |
| Access | API |
| Model identifiers | mimo-v2-tts |
| Voice controls | Natural-language style directions, local vocal events and singing.Original MiMo V2 TTS announcement |
| Training data scale | Xiaomi reported hundreds of millions of hours of speech.Original MiMo V2 TTS announcement |
Related models
- MiMo V2.5 TTS
Xiaomi · MiMo V2.5 Speech
The standard V2.5 TTS model turns text into speech using Xiaomi's built-in voices.
Launch date not confirmedAPI - MiMo V2.5 TTS VoiceClone
Xiaomi · MiMo V2.5 Speech
VoiceClone generates speech using a voice sampled from audio.
Launch date not confirmedAPI - MiMo V2.5 TTS VoiceDesign
Xiaomi · MiMo V2.5 Speech
VoiceDesign creates a new voice from a written description.
Launch date not confirmedAPI